[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

diff3 problems

From: Ben Collins-Sussman <sussman_at_collab.net>
Date: 2002-03-14 21:47:52 CET

Some of you may have noticed that basic_tests.py #5 and #6 are failing
suddenly. Here's the story.

Those tests perform merges; so obviously, they're most likely to be
affected by our switch to diff3 yesterday.

As I was getting ready to commit the "use diff3" changeset, I
discovered that test #5 was suddenly receiving an extra final newline
in the merged text -- a newline that it wasn't expecting. I just
shrugged and assumed that it was some tiny nuance of using diff3
instead of diff+patch. So I changed the test to expect the extra
final newline, and made my commit.

Now it seems that gstein, kfogel, and others are getting test 5
*failures* for exactly this reason: the test is expecting an extra
final newline, but diff3 isn't producing it!

My fear is that this is another Linux/BSD discrepancy.

  --> The back-story is that a long, long time ago we discovered that
      FreeBSD's version of 'GNU diff' was actually a *hacked* version.
      The FreeBSD team removed a GNU-specific feature that allowed the
      program to operate on textfiles that had no final newlines. I'm
      not sure why they did this, but it screwed up our diff+patch
      merging process on FreeBSD. That's why we had an autoconf test
      to look for a "non-broken" GNU diff program.

Of course, I removed that autoconf test yesterday, since we no longer
needed that specific diff+patch behavior. The great Irony is that now
I'm suspicious that we're experiencing the *exact* same problem with
diff3!

We need to examine the exact 3 fulltexts are being merged by diff3 in
basic_test #5, and then see if we can reproduce different behaviors on
different operating systems. Anyone want to look into this? If not,
I'll probably take a look soon. In the meantime, we might want to
change test #5's expectations back to *not* expecting the extra
newline.

(As for test #6, I'm not sure what the deal is -- the actual conflict
markers look very different than the 'expected' ones that I recently
pasted into the test from my FreeBSD box. This also needs
investigation.)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Mar 14 21:48:44 2002

This is an archived mail posted to the Subversion Dev mailing list.